EU's AI Act Transparency Rules Take Effect: Fines up to $17M for Unlabeled AI Content

The EU is moving from declarations to enforcement. This week (August 2), new transparency and labeling rules under the EU AI Act took effect. Companies that develop AI systems (providers) must now disclose to users when they're interacting with AI rather than a human, and must label AI-generated content — image, video, audio, or text — that could be mistaken for real. Companies that deploy these systems (deployers), such as platforms and social networks, face similar obligations, and some companies, like Meta, are classified under both roles. Non-compliance can result in fines of up to €15 million (about $17.2 million) or 3% of a company's global annual revenue. New systems are subject to the law immediately, while existing systems have a grace period until December 2, 2026 to comply. For small and medium businesses using AI tools for marketing content, images, or campaigns, this is a clear signal that now is the time to review labeling and transparency practices. Source: The Verge

Meanwhile, the White House is hosting a meeting today with leading AI companies — OpenAI, Anthropic, and Google — to review a new voluntary model-testing framework. This is a direct continuation of weeks of reports about AI agent security incidents and public calls from industry leaders for increased regulation. The meeting marks a shift from general discussions about the 'pace' of AI development to more concrete steps from the U.S. government. Source: The Verge / CNBC

On the model front, Chinese tech giant Alibaba unveiled Qwen3.8-Max — a massive new model (2.4 trillion parameters) that the company claims outperforms OpenAI's GPT-5.6 Sol Max and Anthropic's Fable 5 on agentic computer-use benchmarks. The model is aggressively priced at about $8 per million tokens, less than a third of Claude Opus 5's price. Alibaba said open weights will be released next week, though the license terms haven't been published yet — a detail that will determine whether businesses can freely adopt the model. Source: VentureBeat

OpenAI launched GPT Live — a continuous voice interaction capability for ChatGPT that enables more natural, flowing voice conversations. The tool continues the trend of making AI assistants more accessible and natural for both consumers and businesses. Source: OpenAI

Finally, Asana revealed a new architecture that lets AI agents 'share memory' across an entire company — so an agent that learned from one project can apply that knowledge to another — while carefully ensuring sensitive information, like a confidential M&A deal, doesn't leak to employees who aren't authorized to know about it. It's a concrete solution to a problem that has come up repeatedly in recent weeks: how to let AI agents learn and improve without compromising privacy and corporate data security. Source: VentureBeat